ElephantBench
Models answer 1,094 closed-book questions about facts with multiple verified but conflicting accounts. A model judge assigns complete, partial, or failed recall.
The benchmark reports recall rates and completeness. Sources and tools are withheld during answering.
- Released
- August 2026
- Kind
- Short factual questions
- Level
- Expert
- Scored
- Graded by a model
- Answers
- Short answer
- Tasks
- All public
- Measures
- Capability
- Language
- English
- Interaction
- Single turn
- Refresh
- One fixed set
- Run by
- Anyone
- Strongest model tested
- GPT-5.6 Sol #14 on the capability index
- Last model added
- July 2026
- Scores refreshed
- November 18th, 2026
In the authors’ words
Factual question answering (QA) typically assumes a single canonical answer, obscuring whether large language models (LLMs) retain divergent accounts of long-tail facts. To address this gap, we introduce ElephantBench, a closed-book knowledge probe comprising 1,094 questions generated through an auditable graph-based pipeline. The pipeline retrieves related documents from a low-exposure web corpus, identifies naturally occurring disagreements, and converts them into multi-account QA records. Each answer is verified against the originating documents and authoritative public web sources and is then reviewed by human annotators. Across 32 models, even the strongest model recovers both accounts on only 52.4% of questions, while on nearly all remaining questions it recalls one account but omits the other. Scaling model size and inference-time reasoning improve recall but do not eliminate this incompleteness. Corpus analysis further shows that exposure imbalance favors the dominant account, whereas greater minority-side exposure is associated with more complete recall. These findings establish ElephantBench as a reproducible knowledge probe for diagnosing epistemic myopia in parametric memory. More broadly, our graph-based benchmark construction pipeline provides an efficient and scalable way to turn long-tail corpora into source-traceable knowledge probes, supporting efforts to evaluate and advance the epistemic rigour of next-generation LLMs. Code is available at https://github.com/Tencent/ElephantBench.
Kimi K3 leads with 52.4% among models released by July 2026 Kimi K3 leads with 52.4% among models released by July 2026
Best score per entry from the benchmark's own leaderboard, read on November 18th, 2026. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. This view shows only models we track, each at its best.
The newest model with a score here came out in July 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 15 more
- 52.4 Kimi K3
- 50.4
Gemini 3.1 Pro
- 50.2 GPT-5.5
- 48.6
Gemini 3.5 Flash
- 45.0 GPT-5.6 Sol
- 44.1 Claude Opus 4.8
- 42.6
Tencent Hy3
- 38.1
Nemotron 3 Ultra
- 37.8 Claude Sonnet 5
- 37.3 GLM-5.2
- 36.2
MiniMax-M3
- 32.3
Qwen3.5 397B-A17B
- 32.0
DeepSeek-V4-Pro
- 30.0
DeepSeek-V4-Flash
- 28.2
Qwen3.5-122B-A10B
- 22.5 gpt-oss-120b
- 21.0
Qwen3.5-35B-A3B
- 19.5
Qwen3.5-27B
- 17.0
Qwen 3.6 35B-A3B
- 15.4 Seed-OSS-36B
- 15.0 OLMo-3.1-32B
- 13.7
Gemma 4 26B A4B
- 12.3
Qwen3.5-9B
- 11.9
Gemma 4 31B IT
- 10.2
Qwen3.6 27B
- 9.69
Qwen3.5-4B
- 9.14 gpt-oss-20b
- 5.85
Gemma 4 12B
- 5.21 OLMo-3-7B
- 2.74
Gemma 4 E4B
- 1.65
Qwen3.5-2B
- 1.37
Gemma 4 E2B
- 1 Kimi K3 52.4%
- 2
Gemini 3.1 Pro 50.4%
- 3 GPT-5.5 50.2%
- 4
Gemini 3.5 Flash 48.6%
- 5 GPT-5.6 Sol 45%
- 6 Claude Opus 4.8 44.1%
- 7
Tencent Hy3 42.6%
- 8
Nemotron 3 Ultra 38.1%
- 9 Claude Sonnet 5 37.8%
- 10 GLM-5.2 37.3%
- 11
MiniMax-M3 36.2%
- 12
Qwen3.5 397B-A17B 32.3%
- 13
DeepSeek-V4-Pro 32%
- 14
DeepSeek-V4-Flash 30%
- 15
Qwen3.5-122B-A10B 28.2%
- 16 gpt-oss-120b 22.5%
- 17
Qwen3.5-35B-A3B 21%
- 18
Qwen3.5-27B 19.5%
- 19
Qwen 3.6 35B-A3B 17%
- 20 Seed-OSS-36B 15.4%
- 21 OLMo-3.1-32B 15%
- 22
Gemma 4 26B A4B 13.7%
- 23
Qwen3.5-9B 12.3%
- 24
Gemma 4 31B IT 11.9%
- 25
Qwen3.6 27B 10.2%
- 26
Qwen3.5-4B 9.69%
- 27 gpt-oss-20b 9.14%
- 28
Gemma 4 12B 5.85%
- 29 OLMo-3-7B 5.21%
- 30
Gemma 4 E4B 2.74%
- 31
Qwen3.5-2B 1.65%
- 32
Gemma 4 E2B 1.37%
Source: Official ElephantBench leaderboard
- 52.4 Kimi K3
- 50.4
Gemini 3.1 Pro
- 50.2 GPT-5.5
- 48.6
Gemini 3.5 Flash
- 45.0 GPT-5.6 Sol
- 44.1 Claude Opus 4.8
- 42.6
Tencent Hy3
- 38.1
Nemotron 3 Ultra
- 37.8 Claude Sonnet 5
- 37.3 GLM-5.2
- 36.2
MiniMax-M3
- 32.3
Qwen3.5 397B-A17B
- 32.0
DeepSeek-V4-Pro
- 30.0
DeepSeek-V4-Flash
- 28.2
Qwen3.5-122B-A10B
- 22.5 gpt-oss-120b
- 21.0
Qwen3.5-35B-A3B
- 19.5
Qwen3.5-27B
- 17.0
Qwen 3.6 35B-A3B
- 13.7
Gemma 4 26B A4B
- 12.3
Qwen3.5-9B
- 11.9
Gemma 4 31B IT
- 10.2
Qwen3.6 27B
- 9.69
Qwen3.5-4B
- 9.14 gpt-oss-20b
- 5.85
Gemma 4 12B
- 2.74
Gemma 4 E4B
- 1.65
Qwen3.5-2B
- 1.37
Gemma 4 E2B
- 1 Kimi K3 52.4%
- 2
Gemini 3.1 Pro 50.4%
- 3 GPT-5.5 50.2%
- 4
Gemini 3.5 Flash 48.6%
- 5 GPT-5.6 Sol 45%
- 6 Claude Opus 4.8 44.1%
- 7
Tencent Hy3 42.6%
- 8
Nemotron 3 Ultra 38.1%
- 9 Claude Sonnet 5 37.8%
- 10 GLM-5.2 37.3%
- 11
MiniMax-M3 36.2%
- 12
Qwen3.5 397B-A17B 32.3%
- 13
DeepSeek-V4-Pro 32%
- 14
DeepSeek-V4-Flash 30%
- 15
Qwen3.5-122B-A10B 28.2%
- 16 gpt-oss-120b 22.5%
- 17
Qwen3.5-35B-A3B 21%
- 18
Qwen3.5-27B 19.5%
- 19
Qwen 3.6 35B-A3B 17%
- 20
Gemma 4 26B A4B 13.7%
- 21
Qwen3.5-9B 12.3%
- 22
Gemma 4 31B IT 11.9%
- 23
Qwen3.6 27B 10.2%
- 24
Qwen3.5-4B 9.69%
- 25 gpt-oss-20b 9.14%
- 26
Gemma 4 12B 5.85%
- 27
Gemma 4 E4B 2.74%
- 28
Qwen3.5-2B 1.65%
- 29
Gemma 4 E2B 1.37%
Source: Official ElephantBench leaderboard
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Kimi K3 | 52.4% | Jul 2026 | Official leaderboard | |
| 2 | | 50.4% | Feb 2026 | Official leaderboard | |
| 3 | GPT-5.5 | 50.2% | Apr 2026 | Official leaderboard | |
| 4 | | 48.6% | May 2026 | Official leaderboard | |
| 5 | GPT-5.6 Sol | 45% | Jul 2026 | Official leaderboard | |
| 6 | Claude Opus 4.8 | 44.1% | May 2026 | Official leaderboard | |
| 7 | | 42.6% | Jul 2026 | Official leaderboard | |
| 8 | | 38.1% | Jun 2026 | Official leaderboard | |
| 9 | Claude Sonnet 5 | 37.8% | Jun 2026 | Official leaderboard | |
| 10 | GLM-5.2 | 37.3% | Jun 2026 | Official leaderboard | |
| 11 | | 36.2% | Jun 2026 | Official leaderboard | |
| 12 | | 32.3% | Feb 2026 | Official leaderboard | |
| 13 | | 32% | Apr 2026 | Official leaderboard | |
| 14 | | 30% | Apr 2026 | Official leaderboard | |
| 15 | | 28.2% | Feb 2026 | Official leaderboard | |
| 16 | gpt-oss-120b | 22.5% | Aug 2025 | Official leaderboard | |
| 17 | | 21% | Feb 2026 | Official leaderboard | |
| 18 | | 19.5% | Feb 2026 | Official leaderboard | |
| 19 | | 17% | Apr 2026 | Official leaderboard | |
| 20 | Seed-OSS-36B | 15.4% | Official leaderboard | ||
| 21 | OLMo-3.1-32B | 15% | Official leaderboard | ||
| 22 | | 13.7% | Apr 2026 | Official leaderboard | |
| 23 | | 12.3% | Feb 2026 | Official leaderboard | |
| 24 | | 11.9% | Apr 2026 | Official leaderboard | |
| 25 | | 10.2% | Apr 2026 | Official leaderboard | |
| 26 | | 9.69% | Feb 2026 | Official leaderboard | |
| 27 | gpt-oss-20b | 9.14% | Aug 2025 | Official leaderboard | |
| 28 | | 5.85% | Jun 2026 | Official leaderboard | |
| 29 | OLMo-3-7B | 5.21% | Official leaderboard | ||
| 30 | | 2.74% | Mar 2026 | Official leaderboard | |
| 31 | | 1.65% | Feb 2026 | Official leaderboard | |
| 32 | | 1.37% | Mar 2026 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Kimi K3 | 52.4% | Jul 2026 | Official leaderboard | |
| 2 | | 50.4% | Feb 2026 | Official leaderboard | |
| 3 | GPT-5.5 | 50.2% | Apr 2026 | Official leaderboard | |
| 4 | | 48.6% | May 2026 | Official leaderboard | |
| 5 | GPT-5.6 Sol | 45% | Jul 2026 | Official leaderboard | |
| 6 | Claude Opus 4.8 | 44.1% | May 2026 | Official leaderboard | |
| 7 | | 42.6% | Jul 2026 | Official leaderboard | |
| 8 | | 38.1% | Jun 2026 | Official leaderboard | |
| 9 | Claude Sonnet 5 | 37.8% | Jun 2026 | Official leaderboard | |
| 10 | GLM-5.2 | 37.3% | Jun 2026 | Official leaderboard | |
| 11 | | 36.2% | Jun 2026 | Official leaderboard | |
| 12 | | 32.3% | Feb 2026 | Official leaderboard | |
| 13 | | 32% | Apr 2026 | Official leaderboard | |
| 14 | | 30% | Apr 2026 | Official leaderboard | |
| 15 | | 28.2% | Feb 2026 | Official leaderboard | |
| 16 | gpt-oss-120b | 22.5% | Aug 2025 | Official leaderboard | |
| 17 | | 21% | Feb 2026 | Official leaderboard | |
| 18 | | 19.5% | Feb 2026 | Official leaderboard | |
| 19 | | 17% | Apr 2026 | Official leaderboard | |
| 20 | | 13.7% | Apr 2026 | Official leaderboard | |
| 21 | | 12.3% | Feb 2026 | Official leaderboard | |
| 22 | | 11.9% | Apr 2026 | Official leaderboard | |
| 23 | | 10.2% | Apr 2026 | Official leaderboard | |
| 24 | | 9.69% | Feb 2026 | Official leaderboard | |
| 25 | gpt-oss-20b | 9.14% | Aug 2025 | Official leaderboard | |
| 26 | | 5.85% | Jun 2026 | Official leaderboard | |
| 27 | | 2.74% | Mar 2026 | Official leaderboard | |
| 28 | | 1.65% | Feb 2026 | Official leaderboard | |
| 29 | | 1.37% | Mar 2026 | Official leaderboard |
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official ElephantBench leaderboard
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official ElephantBench leaderboard
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
More in Factuality
- SimpleVQA In use Models answer 2,025 short factual questions about images across 9 tasks and 9 topics. An LLM judge grades answers against fixed references.
Qwen3.7-Plus 81.7%
- HallusionBench In use Models answer 1,129 yes or no questions about 346 images, built in pairs to expose language hallucination and visual illusion. Scored by accuracy.
Keye-VL-2.0-30B-A3B 73.5%
- SimpleQA Verified In use 1,000 human-verified, de-duplicated short-answer factual questions measuring parametric knowledge.
Gemini 3.1 Pro 77.5%
- AA-Omniscience v1.1 In use Models answer 3,000 knowledge questions in six domains without tools and may abstain. The index adds correct answers and subtracts wrong ones.
Inkling 2 Small 16.1
- GaRAGe In use Models write long-form RAG answers to 2,366 questions with human-annotated grounding passages. Scoring checks whether answers use the relevant passages.
Amazon Nova Pro 60.7%
- Kimi Hallu Bench In use Models answer hard prompts drawn from practical work tasks. Fact-checking agents score the share of claims that are hallucinated. GPT-6 Astra 4.20%
Data audit
All 10 checks pass. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 4 pass 1 note
- Pass Independent runs Scores from the official leaderboard (32 entries).
- Pass Models scored 29 models we track, plus 3 other entries on its leaderboard.
- Pass Newest score The newest is from July 2026, for Kimi K3.
- Note Frontier models None of the top 10 has a score. The strongest tested is GPT-5.6 Sol (#14).
- Pass Headroom The best is 52.4%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on tencent.github.io: 29 rows for models we track and 3 other entries at the last read.
- Pass Last read November 18th, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.